Gopinath Sahoo

Time series with missing values are highly prevalent in equity markets and remain one of the significant threats to reliable stock trend forecasting. The majority of the interseries structure that could lower predictive uncertainty and lessen overfitting is ignored by traditional imputation-aware models, which mostly rely on intra-series learning. In the current study, we propose a deep learning framework incorporating feature enhancement, feature compression, temporal encoding, and Seq2Seq modeling for spatial-temporal learned embedding trend classification. We first expand raw price data by using a large set of technical indicators. Features are then mapped into MTF and GAF-based images to retain temporal dependencies in a spatially interpretable form, placing strong emphasis on image-driven representation learning. Further efficiency in representation is examined by considering both high-dimensional feature spaces and t-SNE-compressed embeddings. These image representations are passed through a Vision Transformer to extract contextual spatial relations, followed by a Seq2Seq LSTM architecture that models temporal evolution and generates discriminative embeddings. A Fully Connected Neural Network performs the final trend classification while SMOTE tackles the issue of class imbalance. The proposed framework is evaluated on twenty stocks across seven major Indian market sectors. Results show that the proposed pipeline can achieve an accuracy of over 80% in trend prediction. The comparative findings also reveal that MTF and GAF have different predictive natures under the compressed and uncompressed feature space, shedding light on the role of image-based representation and compression in deep recurrent models for financial forecasting.

Let G be a graph on n vertices and A(G), L(G), and |L|(G) be the adjacency matrix, Laplacian matrix and signless Laplacian matrix of G, respectively. The paper is essentially a survey of known results about the spectra of the adjacency, Laplacian and signless Laplacian matrix of graphs resulting from various graph operations with special emphasis on corona and graph products. In most cases, we have described the eigenvalues of the resulting graphs along with an explicit description of the structure of the corresponding eigenvectors.

In this article, we give a complete description of the eigenvalues and the corresponding eigenvectors of the distance matrix and the distance signless Laplacian matrix of corona of two graphs and (denoted by ), when is connected transmission regular and is regular. Further, when is connected transmission regular, we give a complete description of the eigenvalues and the eigenvectors of the distance Laplacian matrix of for any graph . As an application, we find families of distance, distance Laplacian and distance signless Laplacian cospectral nonisomorphic graphs.